摘要
人造物体通常具有非常稳定的形状特征,这种几何上较为持久和稳定的属性为物体识别提供了证据,并且它们相对于外观特征、颜色特征或灰度梯度特征等更具有稳定性和分辨力。基于形状特征对物体进行识别的难点在于物体的颜色、光照、尺寸、位置、姿态和背景干扰总是在不断变化的,且不可能事先预测到所有可能的环境。这种物体本身和环境的多样性使得基于几何形状的物体识别成为一个非常具有挑战性的难题。通过定义一种基于形状模板、对边缘线段化之后的图像进行几何证据筛选、收集和组合判断的方法,实现从背景环境中精确找到目标物体,并能够指出组成物体的线段的语义属性。该方法的实质是解一个全局最优的组合优化问题。虽然全局最优组合优化问题看似复杂度很高,但它无需定义复杂的特征向量,无需高代价的样本训练过程,具有非常好的泛化能力和环境适应性,并且能够使用几何特征来提高组合效率,具有更可靠的认知心理学依据。此外,几何证据收集过程简洁明了及具有普适性的特点使其表现出极大的应用前景。实验结果证明该方法在应对环境变化、不变性识别、精确指出物体的几何构造、搜索效率与计算量等方面表现出显著的优势。这一尝试有助于理清发生在物体识别过程中的一些带有普遍性的加工环节。
Artificial objects usually have very stable shape feature, which has persistent and stable properties in geome- try and provides evidence for object recognition- Besides, shape feature is more stable and more discriminative than ap- pearance feature, color feature, gray feature, and gradient feature. The difficulty of object recognition based on shape fea- ture is that objects may change in color, lighting, size, position, pose, and background interference. And we are unable to predict all possible conditions. The variety of objects and conditions make object recognition based on geometric features be a very challenging problem. This paper gave a method based on shape templates, which performs geometric evidence selection, collection,and combination discrimination for the edge segments of images, to find out the target object accu- rately from background, and it is able to point out the semantic attribute for each line segment of the target object. In es- sence, the method is solving a global optimal combinatorial optimization problem. Although the complexity of the global optimal combinatorial optimization problem seems to be very bigh, it is no need to define the complex feature vector and there＇s no need for a high price training process. It has very good generalization ability, environmental adaptability, and more solid basis of cognitive psychology. The geometric evidence collection process, which is simple and universal, shows a great application prospect for this method. The experimental results prove that the method shows great advan- tages in response to the changes in the environment, invariant recognition, pinpointing the geometry of objects, search efficiency, calculation, and etc. This attempt contributes to understanding some universal processing during the process of object recognition.
作者
唐福宇
危辉
TANG Fu-yu WEI Hui(Shanghai Key Laboratory of Data Science, Department of Computer Science, Fudan University,Shanghai 201203,China)
出处
《计算机科学》
CSCD
北大核心
2016年第10期304-311,316,共9页
Computer Science
基金
国家自然科学基金(61375122)
973项目(2010CB327900)资助
关键词
模板法
物体检测
证据积累推理
Template-based method, Object detection, Evidence accumulation reasoning
